The asymptotic decay of the magnitude of the Fourier transform of a function appears always to be determined by its continuity properties as follows, with examples given in Fig. 1:

  • Continuous integral, -20 dB / decade;
  • Continuous function, -40 dB / decade;
  • Continuous 1st derivative; -60 dB / decade;
  • ...

How can this be expressed mathematically in a rigorous way, and proved?

enter image description here
Figure 1. dB magnitude of the Fourier transform of a function as function of the base-10 logarithm of frequency for (from slowest to fastest decay) a rectangular pulse (turquoise), a single lobe of a cosine (purple), and Hann function (blue).


I suppose it would be sufficient to show that a discontinuous function in the time domain decays as $1/\omega$ in the frequency domain (e.g., $\textrm{rect}\Leftrightarrow\textrm{sinc}$), and then use the fact that each integration in the time domain adds a multiplicative factor of $1/j\omega$ in the frequency domain.

  • $\begingroup$ Yes, or start with a sum of shifted and scaled Dirac deltas and show that its spectrum never settles. Or is periodic if the delays are rational numbers. $\endgroup$ – Olli Niemitalo Aug 7 '19 at 10:01

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.